AI APPLIED TO MAGNETIC RESONANCE

COCIR ARTIFICIAL INTELLIGENCE USE CASE 13

MEDICAL FIELD, OR MEDICAL METHODRadiology / Medical Imaging / Diagnostics / Magnetic Resonance

TYPE Decision support Autonomous decision making

CATEGORY Prevention Detection Diagnosis Treatment Other

DESCRIPTION AIR™ Recon DL is a deep learning-based convolutional neural network designed to intelligently reconstruct a final MR image with high SNR and improved image sharpness.AIR™ Recon DL is not a filter or post-processing technique but rather is embedded directly in the reconstruction pipe-line, where the neural network model is applied to input data to remove noise and ringing artifacts prior to final image formation. This means AIR™ Recon DL can access the full set of acquired source data to generatean image, compared to post DICOM image conversion where important information has already been lost.

AIM / PURPOSE AIR™ Recon DL is a deep learning-based convolutional

neural network designed to intelligently reconstruct a final MR image with high SNR and improved image sharpness.

OUTPUT / RESULTS The result of the AI AIR™ Recon DL’s neural network is an image with high SNR and spatial resolution. With AIR™ Recon DL, radiologists can have higher consistency and quality in the images they interpret. Technologists can ac-quire higher SNR images with shorter scan times than usu-al for this level of SNR. No more compromises or tradeoffs in MRI (SNR vs time vs spatial resolution) compared to conventional reconstruction algorithms.

AI METHODOLOGYDeep learning-based convolutional Neural Network.

INPUT / SIZE OF THE DATAAIR™ Recon DL’s neural network is trained on over 10,000 images using GE’s Edison AI Platform to recognize patterns characteristic of noise and low resolution. The trained network employs a cascade of over 100,000 unique pattern recognitions for noise and low resolution to reconstruct only the ideal object image.The network includes a tunable SNR improvement level (3 different levels) to suit the user’s preference. It also includes an innovative ringing suppression technology that recog-nizes common artifacts like Gibbs ringing and truncation and recasts it into improved image detail.

COCIR, the European Coordination Committee of the Radiological, Electromedical and Healthcare IT Industry

© C

OC

IR D

EC

EM

BE

R 2

020

/

D-NICE

GR

AP

HIC

DE

SIG

N

→

REFERENCE DOCUMENTS / LINKS / PUBLICATIONS1. Sneag, D., Potter, H., Roux, P. A new era of deep-learning

image reconstruction. SIGNA™ Pulse of MR. Autumn: 9-13, 2019.

2. Motosugi, U. Deep-learning-based MR reconstruction designed to address compromise between SNR, scan time and resolution. SIGNA™ Pulse of MR. Spring: 7-9, 2020.

3. Van der Velde, N., Bakker, B., Hassing, C., Wielopolski, P.A., Lebel, R.M., Janich, M.A., Budde, R.P., Hirsch, A. Improvement of Late Gadolinium Enhancement Image Quality Using a Novel, Deep Learning Based, Recon-struction Algorithm and Its Influence on Myocardial Scar Quantification. Proceedings of the RSNA. Abstract SSA22-06. 2019.

4. Bash, S.C., Thomas, M., Fung, M., Lebel, R.M., Tanen-baum, L.N. Deep-Learning Reconstruction Improves Quality of Clinical Brain and Spine MR Imaging. Pro-ceedings of the RSNA. Abstract NR370-SD-MOA1. 2019.

5. Argentieri, E.C., Zochowski, K.C., Potter, H.G., Shin, J., Lebel, R.M., Sneag, D.B. Performance of a Deep Learn-ing-Based MR Reconstruction Algorithm for the Eval-uation of Peripheral Nerves. Proceedings of the RSNA. Abstract SSG08-08. 2019.

6. Villaneuva-Meyer, J., Shin, D., Li, Y., Leon III, P.A., Baner-jee, S., Brau., A.C., Lebel, R.M., Glastonbury, C.M. De-noising MR Images of the Cervical Spine: Multi-Reader Assessment of a Deep Learning Approach. Proceedings of the RSNA. Abstract SSQ15-08. 2019.

7. Quarterman P, Lignelli A, Lebel M, Jambawalikar S Deep Learning Reconstruction Method for Improved Visualization of Hippocampal Anatomical Structures, Proceedings of the ISMRM. Abstract #930, 2020

8. Lee HJ, Kang T, Lebel M, Song JE, Sung-Min Gho. Feasi-bility of myelin water fraction mapping with denoised 2D multi-echo GRE images based on deep learned

reconstruction, Proceedings of the ISMRM. Abstract #1426, 2020

9. Lee HJ, Yeonah Kang, Lebel M, Park MS, Lee J. Feasibility of Deep Learned Reconstruction for Dual-Echo Fast Spin Echo Based T2 Mapping of the Hippocampus and Hippocampal Subfield, Proceedings of the ISMRM. Ab-stract #1889, 2020

10. Quarterman P, Moonis G, Lebel M. Deep-Learning Reconstruction Methods for Improved High-Resolution Imaging of the Lateral Rectus-Superior Rectus Band in a Clinical Setting, Proceedings of the ISMRM. Abstract #1897, 2020

11. Allen T, Bancroft LH, Strigel R, Cashen T et al. Shorten-ing Diagnostic T2w Breast Protocols to Capitalize on the Benefits of a Deep Learning Reconstruction, Proceed-ings of the ISMRM. Abstract #2320, 2020

12. Wang X, Jong Bum Son, Bhosale P et al. High Resolu-tion Prostate T2-weighted MRI with Deep Learning and without an Endorectal Coil, Proceedings of the ISMRM. Abstract #2395, 2020

13. O'Shea A, Guidon A, Lebel R et al. Initial Experience in Abbreviated T2-weighted Prostate MRI Using a Deep Learning Reconstruction, Proceedings of the ISMRM. Abstract #2450, 2020

14. McMillan A, Estowski L, Cashen T et al. Accelerated Whole-body Imaging with Uniform Fat-Suppression and Deep-Learning Reconstruction, Proceedings of the ISMRM. Abstract #2600, 2020

15. Hwang KP, Wang X, Ernst R, et al. Deep learning based image reconstruction for T2-weighted rectal cancer im-aging, Proceedings of the ISMRM. Abstract #2603, 2020

16. Internal ref approval number: JB79841XXa

SOURCE GE Healthcare

COCIR, the European Coordination Committee of the Radiological, Electromedical and Healthcare IT Industry

COCIR ARTIFICIAL INTELLIGENCE USE CASE 13FOLLOWING

→

© C

OC

IR D

EC

EM

BE

R 2

020

/

D-NICE

GR

AP

HIC

DE

SIG

N